Market Requirements

60

61 b. Product;

c. Requested fulfillment point/method;

d. Actual fulfillment point/method;

e. Order point/method;

f. Price;

g. Any incentives/promotions applied;

h. Any service fees applied.

o Cost of sale;

o Product, location and assortment attributes;

o Clickstreams and path to purchase;

o Loyalty and customer data;

o Inventory;

o Promotional tactics (i.e. displays, ads, digital campaigns);

o Sales plans & targets;

o Assortments, range plans, space plans;

o New product information;

o B2B opportunities;

o IoT devices: they fall under both the enterprise and external data categories. The manufacturer or owner of a given asset may have access to sensors within those assets. This sensor data would be considered enterprise data as it may not be available in the public domain;

62

o Customer apps: IKEA, as an example, has an app in which customers can design their dream kitchen and ultimately make the purchase. While consumers agonize over the purchase decisions for weeks, the design activity itself is a demand signal.

• External data: below is a list of sample external data, visibility to which results in a better demand signal.

o Weather;

o New stories;

o Social media;

o Local event calendars & schedules;

o Traffic conditions;

o Public holidays;

o Government assistance distribution schedules;

o Competitor locations;

o Sport teams & stars (i.e. performance, scores, win-loss record);

o Movie reviews/box office sales/show times;

o Music sales, downloads, plays;

o Macroeconomic indicators;

o Ratings & reviews (i.e. Amazon.com, TripAdvisor);

o Housing starts;

o Real estate transactions;

63 o Vehicle registrations;

o Center for Disease Control (CDC) data;

o Commodity prices;

o Census/demographic data;

o Currency exchange rates;

o IoT devices: examples of external IoT signals include traffic sensors, cell phone positions, satellite images.

2. Intelligence: a more timely, holistic view of the world is valuable only if companies are equipped to act in a meaningful way. They must be able to influence a key business metric with the new information available.

Within the demand management domain, there are three high level categories of use cases that emerge in terms of taking an intelligent action:

• Explain prior performance: demand management practitioners and their stakeholders in the organization are constantly challenged to explain changes in selling patterns. The challenge today is that because organizations are blind to many of the factors driving demand at a local level, this process is incredibly manual, typically resides outside of systems and is heavily dependent upon tribal knowledge. JDA customers seek visibility to the layers of demand so that they can articulate why demand is what it is;

• Forecast future demand: below there are the primary use cases for a demand forecast in JDA’s current markets. For the Cognitive Demand initiative, the inventory planning use cases are the priority

64

as this is where JDA Demand is most often used today and thus offers a fast path to customer adoption.

o Inventory planning: the process of determining how much inventory to purchase and where to hold it is heavily dependent upon a forecast. In the past, retailers generally used forecasts at a SKU/Store/Week level to make inventory decisions while manufacturers and distributors typically used forecasts at a SKU/DC/Week or Month level. The emergence of omni-channel buying now means that while companies were happy with weekly forecasts in the past, the demand in some industries is now for hourly forecasts. The next-generation of demand management systems must account for channel and must have the ability to be as granular as hours in the time dimension;

o Labor planning: determining how much and what type of labor should be available in a store or a distribution center relies upon a forecast. However, unlike inventory, these forecasts are not necessarily based upon sales of products. In a distribution center, labor requirements are derived by understanding the inbound and outbound workload. How many trucks will need to be loaded or unloaded during this shift? Are they pallets or slip sheets? How many orders need to be picked? Etc. In stores, the labor plan is equally

65

complex. The number of customers expected to enter the store (foot traffic), the number of items purchased, the quantity of returns are all factors. Labor plans often must be as detailed as 15-minute increments;

o Financial planning: it requires a sales forecast as in important input. The desire is typically that the demand forecast feeding the financial plan be constrained by known limitations such as product availability, labor constraints, physical space, etc. These plans are typically not nearly as granular as inventory plans in terms of product, location or time. However, the expectations for precision are much higher because of the less granular nature;

o Strategic planning: long-term strategic plans are driven by a sales forecast. Like financial plans, these forecasts are typically at a very high-level aggregation.

• Shape future demand: once organizations have a good picture of current market demand, they can employ tactics to shape demand to accomplish the company’s financial goals. Demand shaping activities require many of the same models as demand forecasting, thus these initiatives are closely linked. Below are demand shaping processes that typically require demand modeling.

o Assortment planning: assortment plans are driven by estimates of what consumers will purchase. Forecasts used

66

in assortment planning processes must consider the transferability of demand to properly understand the incremental sales of any specific item to be added or deleted.

In addition to transferability, the notion of leveraging attributes to create a sales forecast is important to this process. Many items involved in the assortment decisions have never been carried in the past and thus sales estimates must be derived from similar items;

o Space planning: when allocating space within a store to product families (macro space) or space within a fixture to items (micro space), an understanding of demand is important. Higher selling products should be placed in premium locations while slower moving products receive less desirable locations. The notion of ‘space elasticity’ may be used to determine the magnitude of sales growth achievable by increasing physical presence. Lastly, consumer segmentation plays a role. Place the children’s cereal on the top shelf where they cannot see or reach it and sales will almost suffer as compared to the item at their eye level. Space decisions are typically made infrequently so demand forecasts can often be in monthly or quarterly buckets. However, because every store is a bit different, the

67

best space plans are those that understand the nuances of the specific store;

o Price & promotion: these plans are often seeded with the sales forecast, compared against the financial plans and then campaigns are planned to bring the sales forecast in line with the financial plan. These processes require the ability to model price elasticity as well as the impact of non-price demand drivers such as advertising, premium positioning in stores, featured locations on the website, etc. Competitor activity is often a major consideration in these processes and may impact the demand forecast significantly. Lastly, halo and cannibalization are major factors in these decisions and thus must be included as factors in the demand models.

3. Learning: after perceiving the world in a more timely, holistic manner and subsequently acting to improve business results, we must be able to learn from the actions and outcomes. Below are sample use cases for a demand management system that learns over time.

• Score input signals: a self-learning system may deduce after some period of time that the weather forecast, for example, in a particular geography is so inaccurate or volatile that it provides no predictive value and thus choose to remove that signal from very specific locations;

68

• Find new signals: a self-learning system can look for additional signals, not currently included in the model and recommend their inclusion;

• Prescriptive resolution levers: the market demands systems based on a “manage by exception” paradigm. In this paradigm, users are alerted to potential problems and asked to take actions. A self-learning system will observe these actions over time and eventually move from simply asking them to take an action to recommending one or more actions based on things they or their peers may have done in the past. In time, the system may move from recommending a resolution to directly taking an action and merely notifying a user.